Boggins AI: Vision Enhancer

A dual (Tensorflow & PyTorch) implementation of SRGAN based on CVPR 2017 paper Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network and ESRGAN+ : Further Improving Enhanced Super-Resolution Generative Adversarial Network

Datasets used to train: CelebA and imdb-wiki

Clone our project using: git clone https://github.com/bobbyrathoree/boggins.git

Create an environment using conda or venv, activate it and run: pip install -r requirements.txt

Train

Create data directory: mkdir data
Unzip the dataset files (all images) into data directory.
Run: python train.py --epochs <desired_epochs> --batch <desired_batch_size>

Note: If epochs and batch are not set, the model trains for 50000 epochs using a batch size of 32.

Test

Run: python test.py --input <path-to-your-image-file>

Note: The file is saved in test_results directory.

For training and testing results, visit the website.